Attadale Partners Data Scientist Interview Questions & Guide 2026

The 60-second brief

As a Data Scientist at Attadale Partners, you will occupy a pivotal role at the intersection of advanced analytics and corporate education. This position is not merely about building models; it is about distilling complex datasets into actionable insights that drive professional development curricula and corporate training strategies. You will be responsible for bridging the gap between raw data and the pedagogical frameworks that help organizations upskill their workforce.

If the team owns experimentation, expect depth past a two-sample test: minimum detectable effect and its roughly inverse-square-root dependence on sample size (holding power, significance level and variance fixed), variance reduction from pre-period covariates, interference between units, and when a sequential design is the right call.

Focus areas

• Turn a vague request into a measurable question
• Pick a randomisation unit that respects interference
• Size an experiment before anyone launches it

Overview

Your contributions will directly influence the efficacy of training programs, ensuring that learning outcomes are measurable and aligned with business objectives. You will work within a fast-paced environment where your ability to communicate findings to non-technical stakeholders is just as critical as your technical proficiency. Whether you are analyzing learner engagement or predicting skill gaps, your work will directly shape the future of corporate education at Attadale Partners.

Because this role involves corporate training, your ability to explain complex data concepts to stakeholders who may not have a technical background is a primary evaluation metric.

The Attadale Partners interview process

Attadale Partners candidates report 5 rounds · ≈ 4-6 weeks. The stages below are what candidates describe, not a published process.

Initial Contact

An extra round usually exists because something is still open after the standard loop: a skill the earlier interviews did not sample, a level decision, or two interviewers who disagreed. It is rarely a rerun of what you already did well. Ask the recruiter who you are meeting, what function they sit in, and how long the session runs. That is an ordinary scheduling question, and the answer changes what you should prepare. What separates a strong candidate here is treating the round as a fresh evaluation with its own bar, rather than assuming earlier performance carries you through or sinks you.

What to demonstrate

• Whether you can answer well on ground the earlier rounds did not cover, without leaning on what you already said to someone else
• Consistency of the facts in your stories: the same sample size, timeframe, team size and scope of your own role as in earlier conversations
• How you handle an unfamiliar format live, including whether you ask what kind of answer is wanted before producing one

How to prepare

• Ask the recruiter for the interviewer's function, the length, and whether to expect a coding surface, a discussion, or a presentation. Preparing for a 30 minute conversation with a partner team is not the same work as preparing for a 60 minute technical block.
• Write out what each earlier round actually covered, then list the two or three areas nobody probed. That gap is the most likely subject of the extra round.
• Re-read the numbers in the project stories you have already told, so a second telling does not quietly contradict the first.

High-Level Screening

Because the format is not fixed, prepare the reasoning rather than the ritual. Nearly every version of this round draws on the same underlying material: a design you can defend, a metric you can define exactly, an analysis whose assumptions you can state out loud. Only the wrapper changes, whether that is a take-home, a live case, a deep dive on past work, or a rough estimate on a whiteboard. Answers rehearsed to fit one shape stall the moment the shape differs. Practise naming the assumption behind a number, then saying how much the conclusion moves if that assumption is wrong.

What to demonstrate

• Whether your justification for a method survives the question 'why not the simpler thing', including when the simpler thing would have worked
• Precision under pressure: what exactly counts as an active user, a conversion or a success, over what window, with what exclusions
• Whether you carry an argument through to a recommendation instead of stopping at a list of tradeoffs

How to prepare

• For each project you plan to mention, write the metric definition in one sentence: numerator, denominator, time window, exclusions. Say it out loud once, because vagueness shows up in speech before it shows up on paper.
• Rehearse the same project at three lengths: two minutes, ten minutes, and a deep dive on one technical decision. Cutting live is harder than it sounds.
• For your headline result, write down what would have had to be true for it to be wrong, and how you ruled that out.

Technical Deep Dive

Before anything else, this round is a reading test. You are given a small schema and a question phrased in business language, and most of the difficulty sits in the gap between them. Who counts as an active user, does a refunded order still count as an order, is that date column an event time or a load time. Weak answers start typing immediately and compute something precise about the wrong population. Strong ones pin the definition in one sentence, name the column that encodes it, then write the query. On a timed assessment with nobody to tell, write the definition in a comment anyway.

What to demonstrate

• Whether an ambiguous term becomes a specific column and filter before any computation happens
• Whether you read the schema for keys and cardinality rather than only for column names
• Whether the result answers the question at the grain it was asked at, per user or per session or per day

How to prepare

• Take three metrics you already use and write down the exact filter and exact grain behind each, then practise stating one of them in a single sentence out loud
• On a schema you have never seen, spend the first minute writing what one row of each table means and which key it is unique on, then predict which joins can duplicate rows
• Rehearse a version where the definition changes halfway through, and edit the query you have instead of starting over

Behavioral Assessment

Most of the weight in this round sits on the disagreement questions. Data work routinely produces an answer someone senior did not want, and the interviewer is trying to learn what you do in that hour. Both failure modes are common: folding as soon as a director pushes back, and treating the pushback as ignorance to be corrected with a better chart. A strong answer usually contains a specific thing the other person knew that you did not, and describes how you found out whether it changed the conclusion.

What to demonstrate

• Whether you can state the other side's argument accurately before you explain why you disagreed
• What you treated as evidence during the disagreement, such as a rerun under their assumption or a holdout check, rather than persuasion technique
• Whether you distinguish being overruled from being wrong, and can give an example of each

How to prepare

• Write out one disagreement where you turned out to be wrong, and say what in the data misled you. Candidates prepare the story where they were right, and the follow-up asks for the other one.
• For your main disagreement story, be ready to say what result would have made you drop your position. If no such result exists, you were not arguing from the data.
• Practise stating the opposing position out loud in one sentence the stakeholder would accept, then continue the story.

Final Decision

A loop is not scored one interview at a time. The people you meet compare notes afterwards, usually in a meeting you are not in, and the outcome turns on what each of them can say about you when asked. That rewards something other than survival: every room needs one specific thing worth repeating, and none of them can contradict another. The common way to lose is to tell the same project four times with different numbers in it, or to be uniformly fine in a way that leaves nobody with anything to argue for.

What to demonstrate

• Whether your account of a project survives being told twice, with the same scale, the same metric definition and the same numbers each time
• Whether each interviewer leaves with one concrete claim they could make on your behalf later, rather than an absence of complaints
• Whether a question you already answered in an earlier room gets the same answer at the same depth, without visible impatience

How to prepare

• Write a one-page fact sheet for your two or three main projects that fixes the numbers you will quote: rows of data, the metric as a single sentence, the effect you measured and how long the work took. Say them aloud from the sheet until they come out identical every time
• For each kind of room you expect, decide the one sentence you want that interviewer repeating in a debrief, then check during the mock that you said it outright instead of implying it
• Rehearse answering the same project question twice in one sitting, the second time as though you had not just answered it, because the thing that needs fixing is the flatness that creeps into a repeated story

Where candidates lose points

Slicing a flat experiment until a segment reaches significance

Testing one metric across twenty segments at a nominal 5% level produces a significant result about two thirds of the time when nothing is happening anywhere, and the segment that surfaces is by construction the one with the most favourable noise. The reported effect in that slice is then badly overstated, because selection on significance conditions the estimate on being large. What makes it dangerous rather than merely wrong is that a post-hoc segment always has a plausible story attached, so it survives the meeting. The controls are declaring the small number of segments of interest before launch, correcting across the ones tested, and treating anything discovered afterwards as a hypothesis that needs its own adequately-powered test rather than a finding.

Watching an experiment daily and stopping when it crosses significance

A fixed-sample test controls type I error at one pre-declared look. Checking repeatedly and stopping at the first p < 0.05 inflates the false positive rate to roughly 0.15 to 0.20 for ten looks, and it rises further with more frequent checks, because the p-value takes a random walk that will eventually dip below the threshold under the null. The usual defences are a fixed horizon declared before launch, group-sequential boundaries such as O'Brien-Fleming that spend alpha across a planned number of looks, or always-valid confidence sequences that are correct under continuous monitoring. Compounding it, the effect size reported conditional on having crossed the threshold is biased away from zero, and the bias is larger the lower the power was, so an underpowered test that 'won' typically overstates the lift it found.

Reaching for a model before the target metric exists

Before naming an algorithm, write down the label, the prediction time, and the action that changes when the score crosses a threshold. If you cannot say what decision the output drives, any modelling choice is guesswork dressed up as method.

Building features from data that postdates the prediction time

Check every feature against the timestamp at which the model would actually score, and drop anything computed from a window that includes or follows the label event. For a forecasting use case, split train and test by time rather than at random, and split by entity when the same entity recurs.

Questions & practice

If your model produces unexpected results, what steps do you take to debug the process?

Medium · machine learning and modelling

Approach

1. Frame the prediction: the label, the moment of prediction, and the action it triggers.
2. Set a baseline first, so any model has something honest to beat.
3. Check what information would not exist at prediction time, and exclude it.

Follow-up

• What would you monitor after launch to know the model is still valid?
• Where could label leakage enter this setup?

Explain the difference between supervised and unsupervised learning in the context of clustering learner profiles.

Medium · machine learning and modelling

Approach

1. Pick an evaluation metric that matches the cost of each error type, not a default.
2. Check what information would not exist at prediction time, and exclude it.
3. Set a baseline first, so any model has something honest to beat.

Follow-up

• Where could label leakage enter this setup?
• How would you choose the decision threshold, and who owns that choice?

Audit a one-day event extract for structural defects

Easy · data quality · late arrival · pandas

You receive a one-day extract of fct_event as a DataFrame with event_id, occurred_at_utc, received_at_utc, visitor_id, user_id, account_id, event_name, is_bot_flagged and surface. Write a function returning one row per data-quality rule with the rule name, the failing row count and the failing share of the extract. Cover at minimum: duplicate event_id, received_at_utc earlier than occurred_at_utc, occurred_at_utc later than the extract's maximum received_at_utc, account_id present while user_id is NULL, and rows whose occurred_at date differs from their received_at date. Do not drop rows; report only.

Approach

1. Compute the extract's own reference clock first: max(received_at_utc). Wall-clock now() is wrong here because the extract may be replayed days later, which would turn every row into a future-dated failure.
2. Express each rule as a boolean Series over the same index so the checks compose, then aggregate with .sum() and divide by len(df). Building a list of (name, mask) pairs keeps the rule set extensible and keeps one code path for counting.
3. For the duplicate rule, decide and state the convention: df.duplicated('event_id', keep=False).sum() counts every member of a duplicated group, df.duplicated('event_id').sum() counts only the surplus copies. Either is defensible; an unstated choice is not. The rest of this item assumes keep=False.
4. Treat received_at < occurred_at as clock skew, not corruption: occurred_at is client-supplied. Separate it from the date-mismatch rule, which is the one that actually breaks a daily metric keyed on occurred_at.
5. Know which rules imply which before you read the counts. A row whose occurred_at exceeds max(received_at_utc) has its own received_at no later than that maximum, so it is necessarily a clock-skew row as well: the future-dated mask is a subset of the skew mask, always. Neither is a subset of the date-mismatch mask, because skew of a few minutes inside one UTC date mismatches nothing.
6. Return a tidy DataFrame sorted by failing_share descending, and add a boolean column saying whether the rule should block publication, so the output is a decision rather than a list of numbers.

Worked solution (20 min)

1. Parse both timestamp columns with utc=True and assert the dtype, since a silently-object column makes every comparison string-wise and wrong.
2. Set ref = df['received_at_utc'].max() and build the five masks against it.
3. Assemble results as pd.DataFrame(rows) with columns rule, failing_rows, failing_share, blocks_publication.
4. Keep the masks addressable (a dict of name to Series) rather than only their sums, so the overlap between rules can be asserted rather than assumed.
5. Verify the function is pure: assert the input frame's shape is unchanged after the call.

Expected result: A DataFrame with one row per rule, failing_share equal to failing_rows divided by len(df) for every row, and the input frame returned unmodified. The rules overlap rather than partition the extract, and one containment is structural: every future-dated row is also a clock-skew row. The counts therefore must not be summed or presented as a total.

Verify it yourself

• Plant violations deliberately and assert an exact expected vector rather than assuming one hit per rule. On a 10-row frame carrying two rows that share an event_id, one same-date skew row, one row whose occurred_at sits an hour past max(received_at_utc), one row with account_id set and user_id NULL, and one row whose occurred_at falls on the previous UTC date: duplicate_event_id is 2 under the keep=False convention, clock_skew is 2 (the planted skew row plus the future-dated row, which cannot avoid tripping it), future_dated is 1, orphan_account is 1, date_mismatch is 1.
• Assert the containment directly rather than by counting: (future_dated_mask & ~clock_skew_mask).sum() is 0. A non-zero value means the reference clock was not max(received_at_utc).
• Feed a clean frame and confirm all counts are 0 and no rule errors on empty masks.
• Confirm len(input) is identical before and after, since the deliverable is a report and not a cleaned frame.

Follow-up

• The date-mismatch count is 2.1 percent on this extract. What late-arrival rule would you write for a daily metric, and how many days would you hold the number open?
• Duplicate event_id values appear only on the 'core_action_completed' event. What upstream cause would you check before deduplicating?
• Which of these rules should fire an alert at the pipeline, and which should only appear in a weekly review?

Paying accounts with no active seat in 28 days

Medium · anti-join · null-semantics · retention-risk

dim_account holds account_id, account_type, lifecycle_status, seats_licensed. fct_event holds account_id, user_id, occurred_at_utc, is_core_action, and its account_id is NULL for every signed-out and pre-signup event. Find accounts with lifecycle_status = 'active' and account_type <> 'internal' that had no distinct user complete a core action in the trailing 28 days. Return account_id, seats_licensed and days since that account's most recent core action, with NULL where the account has never emitted one. Order by seats_licensed descending.

Approach

1. Build the recent-activity set first: fct_event rows with is_core_action = TRUE, occurred_at_utc >= now() - interval '28 days', and an explicit account_id IS NOT NULL. Making the NULL exclusion explicit in the CTE is what lets you reason about the anti-join afterwards.
2. Express the exclusion with NOT EXISTS (correlated on account_id) or a LEFT JOIN with an IS NULL guard. Do not use NOT IN against this column: it is nullable, and SQL's three-valued logic turns the whole predicate UNKNOWN, returning zero rows.
3. Compute last-seen separately as MAX(occurred_at_utc) per account over all history, LEFT JOINed on, so an account that has never emitted a core action (NULL) is distinguishable from one that went quiet six weeks ago. Those two cases have different causes and different owners.
4. Rank by seats_licensed, or better by the account's current mrr_cents_constant_fx if you are allowed the subscription table, because a silent fifty-seat account is a renewal conversation and a silent one-seat account is noise.
5. Before shipping, check whether the never-seen group is a cluster by signup date or surface. A block of accounts with no events at all is usually an instrumentation gap, not a set of customers who stopped using the product.

Worked solution (25 min)

1. Count NULL account_id rows in fct_event over the window so you know the trap is live in this data rather than hypothetical.
2. Write the active-account spine and the 28-day activity CTE.
3. Write the anti-join with NOT EXISTS, then deliberately run the NOT IN version and record that it returns zero rows.
4. Add the all-time MAX(occurred_at_utc) LEFT JOIN and derive days_since as a date difference.
5. Split the output into 'quiet' and 'never seen' and eyeball the never-seen group for a shared signup window or surface.

Expected result: One row per silent active account with seats_licensed and days_since_last_core_action, ordered by seats_licensed descending. Accounts that never emitted a core action show NULL and are reported as a separate group rather than as the worst churn risks.

Verify it yourself

• Silent accounts plus accounts with at least one active user equals the total count of active non-internal accounts.
• The NOT IN variant returns zero rows, which is the demonstration that a NULL is present.
• No account_id appears twice in the output.
• Every days_since_last_core_action is either NULL or greater than 28.

Follow-up

• How would you distinguish a genuinely idle account from one whose events lost their account_id after an instrumentation change?
• Would you count on fct_event.account_id or resolve user_id through dim_user instead, and what does each choice miss?
• Licensed-seat utilisation is the continuous version of this. How would you turn this boolean into that ratio?

Read an experiment from first exposure, not assignment

Hard · experiments · window-functions · srm

fct_experiment_exposure holds experiment_id, unit_type, unit_id, variant, user_id, assigned_at_utc, first_exposed_at_utc, is_in_analysis_population and planned_end_utc. fct_event holds user_id, occurred_at_utc, is_core_action, and carries events up to a known data cut, :data_cut_utc. For one experiment randomised on unit_type = 'user', return per variant: exposed units, units with at least one core action in the seven days after that unit's own first exposure, the rate, and the variant share of exposed units. Only units whose seven-day window has fully elapsed as of the data cut belong in the readout. Units appearing under more than one variant are excluded from both arms and counted separately.

Approach

1. Run the contamination pass as an aggregate, not a window: SELECT unit_id FROM fct_experiment_exposure WHERE experiment_id = :exp GROUP BY unit_id HAVING COUNT(DISTINCT variant) > 1, then anti-join it away. PostgreSQL rejects COUNT(DISTINCT variant) OVER (PARTITION BY unit_id) outright, since DISTINCT is not implemented for window functions; if you want the test inline, MIN(variant) OVER (PARTITION BY unit_id) <> MAX(variant) OVER (PARTITION BY unit_id) is the equivalent that does run.
2. Do not resolve contamination by keeping the earliest variant. A unit that saw both arms carries treatment from both, so assigning it to either one biases that arm.
3. Define the population as is_in_analysis_population = TRUE AND unit_type = 'user' AND first_exposed_at_utc < planned_end_utc AND first_exposed_at_utc + interval '7 days' <= :data_cut_utc. The horizon filter is what makes the readout reproducible next week instead of drifting with every re-run; the data-cut filter is the one that actually buys seven days of follow-up, since a unit exposed an hour before the horizon otherwise contributes an hour of observation to a seven-day rate.
4. Measure the outcome on a per-unit relative window: LEFT JOIN fct_event on user_id with is_core_action = TRUE and occurred_at_utc in [first_exposed_at_utc, first_exposed_at_utc + interval '7 days'). LEFT JOIN so units with no outcome stay in the denominator at zero rather than being deleted by an inner join.
5. Check the sample ratio before reading the effect: variant share of exposed units against the intended split, tested as a binomial. Run it on the truncated population as well as on the full exposed set, because if one arm exposes later on average the data-cut filter removes more of that arm and can manufacture a ratio mismatch the randomisation did not have. A mismatch on the full set means the exposure data is not a valid randomisation and invalidates the readout rather than being a footnote under it.
6. Report the per-variant rate, the absolute difference, and the fact that the variance unit is unit_id. That is straightforward here only because the grain is already one row per user; a per-session outcome under user randomisation would need a delta-method or bootstrap standard error instead.

Follow-up

• Some units were assigned days before they were exposed. What does analysing the assigned set instead do to the estimated effect, and in which direction?
• The split is 51/49 on 400,000 exposed units. Do you read the result?
• The treatment arm exposes on average two days later than control. What does that do to a fixed calendar outcome window, and which arm does it favour?

How do you prioritize competing data requests from different departments within the organization?

Medium · product sense and metrics

Approach

1. Restate the decision this analysis has to support, and who acts on the answer.
2. State what result would change your recommendation, so the answer is falsifiable.
3. Decompose the metric into the rates that drive it, and say which one you would check first.

Follow-up

• How would you detect that the metric is being gamed rather than genuinely improving?
• Which segment would you cut first, and what would that rule out?

A client reports that their training completion rates have dropped; how would you investigate the root cause?

Medium · product sense and metrics

Approach

1. State what result would change your recommendation, so the answer is falsifiable.
2. Decompose the metric into the rates that drive it, and say which one you would check first.
3. Fix the population and the time window before naming any metric.

Follow-up

• How would you detect that the metric is being gamed rather than genuinely improving?
• Which segment would you cut first, and what would that rule out?

How would you validate the effectiveness of a new training module using historical performance data?

Medium

Approach

1. State your assumptions explicitly before working the problem.
2. Work from the decision backwards to the evidence you would need.
3. Clarify what is being asked and what a complete answer would contain.

Follow-up

• How would you know your answer was wrong?
• What assumption would you test first?

Halve the runtime of a flat test with pre-period data

Medium · cuped · variance reduction · stratification

A test on 28-day core actions per user is powered for six weeks and the team will not wait. You have fct_event (user_id, occurred_at_utc, is_core_action) covering the period before each unit's first_exposed_at_utc in fct_experiment_exposure, and the pre-period count correlates with the in-experiment count at r = 0.6. Apply CUPED, state precisely how much runtime it buys, and handle the units with no pre-period because they signed up after the experiment started.

Approach

1. Define the adjusted outcome: Y_cuped = Y - theta (X - Xbar), with theta = Cov(Y, X) / Var(X) estimated on the pooled sample so theta itself carries no treatment effect. It is unbiased because X is pre-exposure and therefore has equal expectation in both arms.
2. Quantify the gain exactly: Var(Y_cuped) = Var(Y)(1 - r^2) = Var(Y) x 0.64, so 36% of the variance is removed. Required sample scales with variance, so six weeks becomes 3.8 weeks, and at the original sample the MDE shrinks by sqrt(0.64) = 0.8.
3. State the precondition that actually bites: X must be measured strictly before first_exposed_at_utc, not before assigned_at_utc and not before the calendar start of the experiment. Any post-exposure information in the covariate lets the treatment effect leak in and biases the estimate rather than merely failing to help.
4. Handle missing pre-periods as a stratum, never by dropping rows. Dropping changes the population; setting X to the pooled mean leaves those units' adjustment at zero, which is unbiased but buys them no reduction, so report the covered share and expect the blended reduction to fall short of 36%.
5. Note that stratification is the same mechanism with a categorical covariate. Post-stratifying on device_type and referrer_channel removes the variance those stratum means explain, and it composes with CUPED provided the strata are also fixed before exposure.

Worked solution (30 min)

1. Build the covariate: for each unit_id in fct_experiment_exposure, count fct_event rows with is_core_action = TRUE and occurred_at_utc in [first_exposed_at_utc - 28 days, first_exposed_at_utc).
2. Estimate theta on the pooled sample as Cov(Y, X) / Var(X), and confirm Xbar is statistically indistinguishable between arms before using it.
3. Form Y_cuped = Y - theta (X - Xbar) and run exactly the test you would have run on Y.
4. For units with no pre-period, set X to the pooled mean so their adjustment is zero, and report what share of the sample that is.
5. Report unadjusted and adjusted estimates side by side, with the interval narrower on the adjusted one.

Expected result: 1 - 0.6^2 = 0.36 of the outcome variance is removed, so the required sample falls to 64% of the original and the six-week plan becomes about 3.8 weeks. At the original sample the MDE shrinks by a factor of 0.8. The blended reduction is smaller in proportion to the share of units with no pre-period, and that share is stated alongside the result.

Verify it yourself

• Adjusted and unadjusted point estimates should be close. CUPED reduces variance, it does not move the effect, so a large shift means X was imbalanced at baseline and needs investigating rather than celebrating.
• The realised standard error ratio should be near sqrt(1 - r^2) using the r observed in this experiment, not the historical r.
• Var(Y_cuped) must be no larger than Var(Y) for the fitted theta. A larger value means theta has the wrong sign or was estimated on the wrong window.

Follow-up

• Here the covariate is the same metric as the outcome. When would you deliberately pick a different one, and what is the risk if that covariate is itself affected by treatment?
• The point estimate moves by 0.4 standard errors when CUPED is applied. Is that reassuring or alarming, and what would you check?
• How does CUPED combine with cluster randomisation on account_id, and at which grain is the covariate built?

Attribute a flow-completion drop to one client build

Medium · release-regression · instrumentation · staged-rollout

Core-flow completion rate on surface = 'ios', defined as distinct flow_instance_id reaching 'flow_completed' within 30 minutes of 'flow_started' with no 'error_shown' on the same id in between, fell from 78% to 71% over four days. A new iOS build began a staged rollout on day one. Using fct_event columns flow_id, flow_instance_id, app_version, event_name, user_id and occurred_at_utc, produce the completion rate per build per day, size the loss in absolute completions, and say whether the fix is a rollback. Treat the possibility that the instrumentation changed rather than the flow.

Approach

1. Cut the rate by app_version and day, each build against its own denominator. A pooled series during a staged rollout is a weighted average whose weights move daily, so it declines in proportion to the rollout share even when the old build is perfectly flat; confirming that proportionality is itself the evidence the regression is build-specific rather than environmental.
2. Test the denominator before believing the rate. Compute flow_started events per distinct user_id on each build: if the new build re-mints flow_instance_id on retry, or fails to carry the id from start to completion, the denominator inflates and the numerator falls with nothing changing for the user.
3. Normalise to a user-level outcome that is immune to the id question, namely completions per distinct user_id per day on each build. This is the number that says whether anyone actually failed to finish.
4. Check the error path: pull error_shown counts and their properties for flow_instance_id values on the new build. A genuine regression produces errors; an instrumentation break produces missing completions with no corresponding error volume, and the two prescribe different fixes.
5. Size the real loss as the per-user completion gap multiplied by daily users on the new build, and give the rollback recommendation conditional on the error evidence rather than on the rate.

Follow-up

• The rollout is at 45% and product wants to go to 100% tomorrow. What do you say, and what would you need by when?
• If the flow_instance_id is genuinely being re-minted, what is the correct historical treatment of the four affected days?
• How would you have caught this on day one instead of day four?

Suggested practice checklist

For a candidate whose interviews will centre on A/B testing, metric movement and causal claims. Design comes before arithmetic, arithmetic before analysis, and the week ends by rehearsing the readout rather than the derivation.

Day 1 — Design one test end to end on paper

• Take a single feature change and write the full design: randomization unit, the exact point of exposure, the primary metric with its grain, guardrails, allocation, planned duration, and the decision rule committed before any data exists.
• Write why the randomization unit must sit at or above the level where treatment can spill over, and give one case where user-level randomization is still contaminated (shared accounts or devices, or two participants in the same marketplace).
• State in advance what you will do if the primary metric is flat while a secondary metric is significant.

Deliverable: A one-page test design with a decision rule written before launch.

Day 2 — Power arithmetic until it is automatic

• Compute required sample size per arm for a binary metric with the normal approximation, n is approximately 2 times (z for alpha/2 plus z for power) squared times p(1 minus p) divided by delta squared, for baselines of 2, 10 and 40 percent at a 5 percent relative lift, and note that for a fixed relative lift the requirement falls as the baseline rises because delta grows proportionally with p.
• Redo the calculation for a continuous metric using variance in place of p(1 minus p), and show why a heavy-tailed quantity such as revenue per user needs either far more traffic or a capped version with a stated cap.
• Convert one of the results into weeks given a weekly eligible traffic figure, then list the two honest ways to shorten it (accept a larger detectable effect, or reduce variance) and write why quietly lowering the power target is a decision to miss more real wins, not a speedup.

Deliverable: A small script or sheet that maps baseline, minimum detectable effect, alpha and power to sample size and weeks, cross-checked against a published calculator.

Day 3 — Variance and the unit-of-analysis problem

• Take a ratio metric whose denominator is not the randomization unit (clicks per session, randomized by user) and compute the standard error twice, once naively at session level and once by the delta method or a user-level bootstrap, then record how much the naive version understates it.
• Implement CUPED on simulated data: choose a pre-period covariate X measured before assignment, estimate theta as Cov(Y, X) divided by Var(X), and analyse Y minus theta times (X minus its mean) in place of Y. Confirm the variance of the adjusted outcome equals the raw variance multiplied by one minus the squared correlation between Y and X, so a correlation of 0.45 removes about 20 percent of the variance and not 80.
• Now run that simulation a few hundred times and confirm the adjusted effect estimate is unbiased for the same effect rather than numerically identical to the raw one. Within any single run the two differ, sometimes by a large fraction of the true effect, because the two arms' pre-period covariate means never coincide exactly in a finite sample; they agree in expectation, which is the property that matters and the one to state out loud.

Deliverable: A notebook showing the adjusted estimator with a measurably smaller variance than the raw one, plus a repeated-simulation table showing the two estimators agreeing on average while differing run by run.

Day 4 — Validity threats you can actually test for

• Run a sample ratio mismatch check as a chi-square goodness-of-fit test against the intended allocation, and write the three causes you would chase first (assignment logged before exposure, an arm-specific redirect or load failure, bot filtering applied asymmetrically).
• Simulate peeking: generate A/A data, test daily at alpha 0.05 across 14 looks, record the inflated false positive rate, then apply an alpha-spending boundary or commit to a fixed horizon and confirm the rate returns to nominal.
• Write how you would separate a novelty effect from a durable lift using the treatment effect plotted against days since first exposure, and what shape would change your recommendation.

Deliverable: One table showing the peeking false positive rate before and after correction, plus a written SRM triage list.

Day 5 — When randomization is not available

• Write the identifying assumption for difference-in-differences (parallel trends in the absence of treatment), then plot pre-period trends for two candidate control groups and justify rejecting one of them.
• Design a switchback test for a change where user-level randomization would leak across participants, choosing a time-block length against the carryover you expect and saying how you would detect carryover in the data.
• List what an interrupted time series or a synthetic control buys you and the one thing neither can rule out: an unobserved shock that coincides with the launch.

Deliverable: A one-page memo recommending a single quasi-experimental design and naming its weakest assumption explicitly.

Day 6 — The readout query

• Write the assignment-to-exposure join that returns exactly one row per unit per experiment, and handle units appearing in both arms by excluding and counting them rather than silently keeping one.
• Compute the per-arm metric, its variance and the relative lift with a confidence interval in SQL, then reproduce the identical numbers in a notebook as a cross-check.
• Add a segment breakdown and write the sentence that keeps it from being p-hacking: segments declared in advance, everything else reported as exploratory and corrected for multiplicity.

Deliverable: A single query that outputs the full readout table, matched to a notebook recomputation.

Day 7 — Present it to someone who will not read the appendix

• Give a 10-minute readout of a real or simulated experiment in the order decision, number, uncertainty, caveat.
• Have your listener ask "can we ship it" in the case where the primary is flat and a guardrail moved, and answer with a recommendation rather than a request for more data.
• Rewrite your opening line so the recommendation lands before any methodology.

Deliverable: A one-page readout whose first line is the recommendation.

Bring numbers, including the ones you are unsure of

An answer without a quantity is hard to interrogate, so interviewers keep probing until they find one. Come with the baseline, the change, the window it was measured over, and how confident you were. If the effect never got measured, say so and say what you would have measured. Fabricated precision is worse than an honest gap.

Describe a time you chose a simpler model over a more complex one; what was your reasoning?

Medium · behavioural and stakeholder questions

Approach

1. Name the disagreement or constraint, and how you resolved it with evidence.
2. Close with what you would do differently, concretely.
3. Quantify the outcome, including what you would not claim credit for.

Follow-up

• What did you decide not to do, and why?
• How did you know the outcome was caused by your change?

Tell me about a time you had to explain a complex technical insight to a non-technical manager.

Medium · behavioural and stakeholder questions

Approach

1. Pick a story where you drove the decision, not one where you observed it.
2. Name the disagreement or constraint, and how you resolved it with evidence.
3. State the situation in two sentences and spend the rest on your reasoning.

Follow-up

• How did you know the outcome was caused by your change?
• What did you decide not to do, and why?

Quantify your own impact without claiming the topline you touched

Hard · self-assessment · attribution · communication

You are writing the impact section of your own review. Over the year you ran four experiments, one of which shipped and three of which were flat; you corrected the definition of gross monthly revenue churn so that cancellation is recognised at period_end_utc; and you built a self-serve funnel dashboard. Weekly active accounts rose 14% over the same period. Your reviewer knows the data well. Write the three impact claims you would defend, stating for each what you contributed, what evidence supports it, and what portion of the outcome you are not claiming.

Approach

1. Recognise what is being probed: whether you apply to your own work the causal standard you would apply to somebody else's roadmap claim. Nearly everyone who would reject 'accounts that do Y retain better' will write 'I drove a 14% increase' without noticing it is the same error with a friendlier subject.
2. Sort the work by the kind of evidence it can carry. The shipped experiment is the only item with a randomised estimate, so it is the only one where an effect size is defensible, and you claim the interval rather than the point estimate.
3. Claim the three flat experiments as decisions prevented and price them. Features not built, or built differently, on evidence, with the engineering weeks reallocated as the number somebody else can verify. A defensible null is a delivered decision and should be written as one.
4. Claim the definition fix as correctness, not as improvement. The old figure was overstated by a specific percentage and appeared in a specific set of recurring documents; the impact is the change it produced in the forecast built on top of it, not a change in churn itself.
5. Claim the dashboard on usage and displacement: distinct weekly users of it, and the ad-hoc request count for six months before against six months after. If the request log does not exist, record the claim as unverified rather than estimating it upward.
6. Disclaim the 14% explicitly and once. State that it cannot be separated from seasonality, other teams' launches and a pricing change, and bound your own contribution from above using the shipped experiment's interval converted into headline units.

Follow-up

• Your shipped experiment's interval was +0.2pp to +1.4pp on activation. How much of the 14% can that account for, and how do you say so without undercutting yourself?
• A peer in the same cycle claims the full 14%. What, if anything, do you do about it?
• If you could only keep two of your three claims, which do you drop, and why that one?